Almost every fashion company can now say it uses AI.
That statement is becoming less useful by the month.
A marketing manager using ChatGPT to rewrite a product description and a retailer whose AI personalizes discovery, assembles outfits and takes actions on behalf of a shopper are both technically “using AI.” Strategically, however, they are doing completely different things.
A better way to assess AI maturity is to ask where AI sits in the business, what decisions it influences and how directly it affects the customer or the P&L.
DRESSX’s 2026 AI Fashion E-Commerce Guide formalizes that progression through three waves: Wave 1: Copilot, Wave 2: Embedded AI and Wave 3: Autonomous AI. DRESSX
The framework is useful because it shifts the conversation away from the number of AI tools a company has purchased.
The important question becomes: what has AI actually changed?
Wave 1: Copilot — AI makes the existing company faster
Wave 1 is where most adoption begins.
A human already has a task. AI accelerates part of it.
The marketing team uses generative AI for campaign concepts or product copy. Customer support uses it to draft responses. Designers generate image variations. Analysts summarize large datasets. Merchandisers ask questions of spreadsheets. SEO teams accelerate first drafts.
This is the copilot model.
The human remains responsible for the workflow and generally makes the final decision.
Shopify’s 2025 merchant survey illustrates how mainstream this layer has become. Seventy-five percent of surveyed store owners reported using AI tools, and content generation was the most common use case at 69%. Shopify
There is genuine value here. If a six-person ecommerce team can produce the output of an eight-person team without degrading quality, the economic effect is meaningful.
But Wave 1 has an important limitation: the shopper may never notice.
A company can use AI to produce every PDP description, campaign email and internal report while leaving the storefront almost identical to what it was several years earlier.
The business is operating faster, but the commerce experience has not fundamentally changed.
That is why the primary metrics in Wave 1 are productivity metrics: hours saved, assets produced, support handling time, cost per creative variant, campaign velocity and employee adoption.
Wave 1 becomes a trap when executives confuse AI activity with AI transformation.
Having 15 subscriptions in an “AI stack” is not a strategy.
A useful test for Wave 1
Imagine turning every AI product in the organization off tomorrow.
Would the customer’s shopping journey look essentially the same?
If the answer is yes, the company is probably still predominantly in Wave 1.
There is nothing wrong with that. The mistake is believing the company is further along simply because internal usage is high.
BCG argues that leading retailers will need to move beyond attaching AI tools to legacy workflows and instead redesign customer propositions, economics, operating models and technology around the capabilities AI creates. BCG
That transition is Wave 2.
Wave 2: Embedded AI — AI becomes part of the shopping experience
Wave 2 begins when AI enters a core customer or commercial decision.
Instead of merely helping someone write the product description, it influences which product the shopper sees.
Instead of helping the merchandising team analyze search queries, it changes the search results in real time.
Instead of creating campaign imagery, it helps the customer visualize a product on herself.
In fashion, Wave 2 includes capabilities such as personalized search, multimodal search, AI styling, outfit generation, virtual try-on, fit and sizing systems, personalized merchandising, inventory-aware shopping assistants and AI-assisted allocation or markdown decisions.
The defining question changes.
Wave 1 asks, “Did AI make us faster?”
Wave 2 asks, “Did AI make the business perform better?”
That means conversion rate, AOV, items per order, return rate, gross margin, sell-through, retention and inventory efficiency begin to matter more than hours saved.
Fashion is unusually suited to Embedded AI
Several of fashion’s hardest customer problems are reasoning problems.
Compatibility asks what goes together. Personalization asks what works for this shopper. Context asks what works for this occasion. Visualization asks what the item might look like on a person rather than the model. Fit asks whether a size or cut is appropriate. Inventory asks whether the recommended product can actually be purchased now.
Traditional ecommerce handles those questions through different components, and in many journeys it leaves the shopper to combine the answers herself.
AI can increasingly bring them together.
BCG describes the retail journey moving toward mission-based discovery, where consumers seek outcomes such as refreshing a wardrobe rather than independently navigating products and categories. Retailers that want to remain direct destinations will need stronger assisted discovery and data-driven personalization. BCG
That is fundamentally a Wave 2 opportunity.
Elara belongs primarily in Wave 2 today
Elara’s commerce product is best understood as Embedded AI.
Its styling interface sits in the fashion storefront and lets customers describe an occasion, vibe or constraint in natural language. The system then assembles a shoppable outfit from the retailer’s inventory. Elara also supports outfit bundling and virtual try-on, while its Shopify onboarding includes catalog ingestion and enrichment. Elara
The distinction from a Wave 1 AI tool is straightforward.
The AI is no longer merely helping an employee.
It is participating directly in the revenue journey.
Its usefulness therefore has to be evaluated commercially.
Elara publishes impact ranges based on benchmarks from comparable AI-merchandising deployments, not mature Elara-specific production studies. That makes controlled testing particularly important. Elara
An onsite AI stylist should not be considered successful because it produces good-looking conversations.
It should be considered successful because it changes a business outcome.
The biggest measurement mistake in Wave 2
Suppose 20% of shoppers interact with an AI stylist.
Is that good?
Not necessarily.
Suppose those shoppers spend 30% longer on the website.
Still unclear.
More time on site could mean stronger engagement, or it could mean the experience makes the shopper work harder.
The critical Wave 2 question is incrementality: what happened because the AI existed that would not otherwise have happened?
That usually requires some form of controlled comparison.
For an AI stylist, the retailer should care about assisted conversion, AOV, units per order, gross margin per session, subsequent returns and repeat behavior. For virtual try-on, it should look at PDP-to-cart behavior, conversion and returns. For AI search, it should examine search exits, zero-result behavior, discovery and revenue generated from search sessions.
The exact metrics depend on the use case.
The principle does not.
Wave 2 AI has to earn its place on the P&L.
Wave 3: Autonomous AI — AI can take action
Wave 3 changes the role of the system again.
A copilot recommends.
Embedded AI participates in a decision.
Autonomous AI can execute a defined action within explicit permissions.
For a consumer fashion agent, that could eventually mean identifying a wardrobe gap, finding suitable products, checking size and budget, evaluating whether the product works with the shopper’s wardrobe, comparing options, waiting for a price condition and completing the transaction after the required approval.
At the retailer level, autonomous systems could eventually monitor inventory, generate merchandising recommendations, adjust defined parameters, initiate campaigns or execute certain operational decisions within agreed guardrails.
The key word is not intelligence.
It is agency.
Autonomy does not mean removing humans
Wave 3 is sometimes described as if the goal were to hand the entire business to an AI system.
That would be a poor operating model.
The useful question is which actions can safely be delegated, under what conditions, with which data, with what approval thresholds and with which audit mechanisms.
Google’s India commerce research shows that consumers are already open to forms of delegated purchasing: 45% of surveyed shoppers said they would be comfortable allowing AI to purchase on their behalf provided they were notified before the final transaction. Google
BCG similarly argues that AI-era retailers will need explicit governance, human-AI decision rights and infrastructure that tracks value rather than activity. BCG
The difference between useful autonomous systems and dangerous autonomous systems will largely be the quality of those boundaries.
Elara’s roadmap points toward Wave 3, but the distinction matters
Elara’s public commerce page currently labels its agentic systems as “coming soon.” It describes future functions including order placement, order management, payments, returns and exchanges within the conversational experience. Elara
That means it would be inaccurate to position the current product as a fully autonomous fashion-commerce agent.
Today, its stronger positioning is Wave 2: embedded styling intelligence.
The roadmap extends toward Wave 3.
That sequence is logical because reliable autonomy should be built on top of domain intelligence. The system first needs to understand the catalog. Then it needs reliable fashion reasoning and personalization. Then it needs transaction permissions and operational guardrails.
Giving a weak recommendation engine purchasing authority does not make it sophisticated.
It makes mistakes more expensive.
Brands can occupy all three waves simultaneously
The framework should not be interpreted as a companywide maturity score.
A fashion retailer’s marketing team may operate in Wave 1, using generative AI for creative production. Its onsite search experience may be Wave 2. A tightly bounded inventory workflow could already contain Wave 3 characteristics.
What matters is the maturity of each use case.
This prevents an executive team from chasing “Wave 3” simply because autonomy sounds more advanced.
The objective is not to progress every process toward full agency.
It is to match the appropriate level of AI to the economic value and risk of the decision.
Most brands should not jump directly to Wave 3
Agentic commerce receives disproportionate attention because it sounds like the most futuristic part of AI.
For many fashion brands, however, the higher-return opportunity today is much simpler: identify one expensive customer decision and make the store materially better at solving it.
That decision could be “What should I wear to this event?” It could be “Which size should I buy?” It could be “What works with this jacket?” or “Which of these 200 dresses actually matches what I described?”
Solving one of those problems extremely well creates useful behavioral data, organizational expertise and customer trust.
Those assets become the foundation for more autonomous systems later.
Skipping that stage usually means putting agency on top of weak intelligence.
The strategic danger is staying in Wave 1 forever
Copilots will become standard infrastructure.
Every brand will be able to generate product copy quickly. Every creative team will be able to produce more variations. Every analyst will summarize reports faster. Every support team will use AI.
Those productivity gains matter.
But once everyone has them, they stop being differentiation.
The more strategic question is what your AI understands or does for the shopper that a competing storefront does not.
That could be better sizing intelligence. Better outfit compatibility. A richer fashion ontology. Persistent understanding of customer taste. Better first-party preference data. Superior mission-based discovery. Eventually, it may be a trusted shopping agent with permission to act.
Those assets are much harder to copy than an AI copywriting workflow.
Wave 1 makes the existing fashion company faster.
Wave 2 changes the fashion shopping experience.
Wave 3 may eventually change who—or what—performs parts of the shopping process.
The brands that understand those differences will make much better AI investments than the brands whose strategy is simply to “use more AI.”